A computational framework integrating multi-omics and machine learning for identifying glycolytic gene markers in breast cancer

Breast cancer remains a leading cause of cancer-related mortality in women worldwide. Metabolic reprogramming, especially enhanced glycolysis, is a hallmark of breast cancer progression and offers promising targets for therapeutic intervention. However, systematic identification of glycolysis-related gene markers at single-cell resolution remains limited.We integrated single-cell RNA sequencing, spatial transcriptomics, and multiple machine learning algorithms to analyze glycolytic heterogeneity in breast cancer. A novel interquartile range (IQR)-based scoring method was developed to quantify glycolytic activity at the single-cell level. Glycolysis-related genes (GRGs) were systematically screened using five machine learning models (RF, Boruta, Lasso, ABESS, and GBM), and their diagnostic and prognostic values were evaluated.We identified significant glycolytic heterogeneity among breast cancer cell subsets, with malignant cells exhibiting the highest glycolytic activity. Seven hub genes, namely PKM, MIF, SOD1, PGK1, APP, LDHB and NUPR1, were consistently identified and validated. These genes showed strong diagnostic potential with high AUC values in ROC analysis, and their elevated expression was confirmed in breast cancer tissues via immunohistochemistry. Spatial and cell communication analyses further revealed distinct metabolic niches and intercellular signaling networks associated with high-glycolytic cells.This study establishes a robust IQR-based glycolysis assessment strategy that overcomes limitations of previous scoring methods. We identified seven core glycolytic regulatory genes that are closely linked to breast cancer progression and poor prognosis. These findings provide novel insights into metabolic reprogramming in breast cancer and offer potential biomarkers for targeted metabolic therapy.